xref: /openbmc/qemu/hw/riscv/virt.c (revision b2580720)
1 /*
2  * QEMU RISC-V VirtIO Board
3  *
4  * Copyright (c) 2017 SiFive, Inc.
5  *
6  * RISC-V machine with 16550a UART and VirtIO MMIO
7  *
8  * This program is free software; you can redistribute it and/or modify it
9  * under the terms and conditions of the GNU General Public License,
10  * version 2 or later, as published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope it will be useful, but WITHOUT
13  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
15  * more details.
16  *
17  * You should have received a copy of the GNU General Public License along with
18  * this program.  If not, see <http://www.gnu.org/licenses/>.
19  */
20 
21 #include "qemu/osdep.h"
22 #include "qemu/units.h"
23 #include "qemu/error-report.h"
24 #include "qemu/guest-random.h"
25 #include "qapi/error.h"
26 #include "hw/boards.h"
27 #include "hw/loader.h"
28 #include "hw/sysbus.h"
29 #include "hw/qdev-properties.h"
30 #include "hw/char/serial.h"
31 #include "target/riscv/cpu.h"
32 #include "hw/core/sysbus-fdt.h"
33 #include "target/riscv/pmu.h"
34 #include "hw/riscv/riscv_hart.h"
35 #include "hw/riscv/virt.h"
36 #include "hw/riscv/boot.h"
37 #include "hw/riscv/numa.h"
38 #include "kvm/kvm_riscv.h"
39 #include "hw/firmware/smbios.h"
40 #include "hw/intc/riscv_aclint.h"
41 #include "hw/intc/riscv_aplic.h"
42 #include "hw/intc/sifive_plic.h"
43 #include "hw/misc/sifive_test.h"
44 #include "hw/platform-bus.h"
45 #include "chardev/char.h"
46 #include "sysemu/device_tree.h"
47 #include "sysemu/sysemu.h"
48 #include "sysemu/tcg.h"
49 #include "sysemu/kvm.h"
50 #include "sysemu/tpm.h"
51 #include "sysemu/qtest.h"
52 #include "hw/pci/pci.h"
53 #include "hw/pci-host/gpex.h"
54 #include "hw/display/ramfb.h"
55 #include "hw/acpi/aml-build.h"
56 #include "qapi/qapi-visit-common.h"
57 #include "hw/virtio/virtio-iommu.h"
58 
59 /* KVM AIA only supports APLIC MSI. APLIC Wired is always emulated by QEMU. */
60 static bool virt_use_kvm_aia(RISCVVirtState *s)
61 {
62     return kvm_irqchip_in_kernel() && s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC;
63 }
64 
65 static bool virt_aclint_allowed(void)
66 {
67     return tcg_enabled() || qtest_enabled();
68 }
69 
70 static const MemMapEntry virt_memmap[] = {
71     [VIRT_DEBUG] =        {        0x0,         0x100 },
72     [VIRT_MROM] =         {     0x1000,        0xf000 },
73     [VIRT_TEST] =         {   0x100000,        0x1000 },
74     [VIRT_RTC] =          {   0x101000,        0x1000 },
75     [VIRT_CLINT] =        {  0x2000000,       0x10000 },
76     [VIRT_ACLINT_SSWI] =  {  0x2F00000,        0x4000 },
77     [VIRT_PCIE_PIO] =     {  0x3000000,       0x10000 },
78     [VIRT_PLATFORM_BUS] = {  0x4000000,     0x2000000 },
79     [VIRT_PLIC] =         {  0xc000000, VIRT_PLIC_SIZE(VIRT_CPUS_MAX * 2) },
80     [VIRT_APLIC_M] =      {  0xc000000, APLIC_SIZE(VIRT_CPUS_MAX) },
81     [VIRT_APLIC_S] =      {  0xd000000, APLIC_SIZE(VIRT_CPUS_MAX) },
82     [VIRT_UART0] =        { 0x10000000,         0x100 },
83     [VIRT_VIRTIO] =       { 0x10001000,        0x1000 },
84     [VIRT_FW_CFG] =       { 0x10100000,          0x18 },
85     [VIRT_FLASH] =        { 0x20000000,     0x4000000 },
86     [VIRT_IMSIC_M] =      { 0x24000000, VIRT_IMSIC_MAX_SIZE },
87     [VIRT_IMSIC_S] =      { 0x28000000, VIRT_IMSIC_MAX_SIZE },
88     [VIRT_PCIE_ECAM] =    { 0x30000000,    0x10000000 },
89     [VIRT_PCIE_MMIO] =    { 0x40000000,    0x40000000 },
90     [VIRT_DRAM] =         { 0x80000000,           0x0 },
91 };
92 
93 /* PCIe high mmio is fixed for RV32 */
94 #define VIRT32_HIGH_PCIE_MMIO_BASE  0x300000000ULL
95 #define VIRT32_HIGH_PCIE_MMIO_SIZE  (4 * GiB)
96 
97 /* PCIe high mmio for RV64, size is fixed but base depends on top of RAM */
98 #define VIRT64_HIGH_PCIE_MMIO_SIZE  (16 * GiB)
99 
100 static MemMapEntry virt_high_pcie_memmap;
101 
102 #define VIRT_FLASH_SECTOR_SIZE (256 * KiB)
103 
104 static PFlashCFI01 *virt_flash_create1(RISCVVirtState *s,
105                                        const char *name,
106                                        const char *alias_prop_name)
107 {
108     /*
109      * Create a single flash device.  We use the same parameters as
110      * the flash devices on the ARM virt board.
111      */
112     DeviceState *dev = qdev_new(TYPE_PFLASH_CFI01);
113 
114     qdev_prop_set_uint64(dev, "sector-length", VIRT_FLASH_SECTOR_SIZE);
115     qdev_prop_set_uint8(dev, "width", 4);
116     qdev_prop_set_uint8(dev, "device-width", 2);
117     qdev_prop_set_bit(dev, "big-endian", false);
118     qdev_prop_set_uint16(dev, "id0", 0x89);
119     qdev_prop_set_uint16(dev, "id1", 0x18);
120     qdev_prop_set_uint16(dev, "id2", 0x00);
121     qdev_prop_set_uint16(dev, "id3", 0x00);
122     qdev_prop_set_string(dev, "name", name);
123 
124     object_property_add_child(OBJECT(s), name, OBJECT(dev));
125     object_property_add_alias(OBJECT(s), alias_prop_name,
126                               OBJECT(dev), "drive");
127 
128     return PFLASH_CFI01(dev);
129 }
130 
131 static void virt_flash_create(RISCVVirtState *s)
132 {
133     s->flash[0] = virt_flash_create1(s, "virt.flash0", "pflash0");
134     s->flash[1] = virt_flash_create1(s, "virt.flash1", "pflash1");
135 }
136 
137 static void virt_flash_map1(PFlashCFI01 *flash,
138                             hwaddr base, hwaddr size,
139                             MemoryRegion *sysmem)
140 {
141     DeviceState *dev = DEVICE(flash);
142 
143     assert(QEMU_IS_ALIGNED(size, VIRT_FLASH_SECTOR_SIZE));
144     assert(size / VIRT_FLASH_SECTOR_SIZE <= UINT32_MAX);
145     qdev_prop_set_uint32(dev, "num-blocks", size / VIRT_FLASH_SECTOR_SIZE);
146     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
147 
148     memory_region_add_subregion(sysmem, base,
149                                 sysbus_mmio_get_region(SYS_BUS_DEVICE(dev),
150                                                        0));
151 }
152 
153 static void virt_flash_map(RISCVVirtState *s,
154                            MemoryRegion *sysmem)
155 {
156     hwaddr flashsize = virt_memmap[VIRT_FLASH].size / 2;
157     hwaddr flashbase = virt_memmap[VIRT_FLASH].base;
158 
159     virt_flash_map1(s->flash[0], flashbase, flashsize,
160                     sysmem);
161     virt_flash_map1(s->flash[1], flashbase + flashsize, flashsize,
162                     sysmem);
163 }
164 
165 static void create_pcie_irq_map(RISCVVirtState *s, void *fdt, char *nodename,
166                                 uint32_t irqchip_phandle)
167 {
168     int pin, dev;
169     uint32_t irq_map_stride = 0;
170     uint32_t full_irq_map[GPEX_NUM_IRQS * GPEX_NUM_IRQS *
171                           FDT_MAX_INT_MAP_WIDTH] = {};
172     uint32_t *irq_map = full_irq_map;
173 
174     /* This code creates a standard swizzle of interrupts such that
175      * each device's first interrupt is based on it's PCI_SLOT number.
176      * (See pci_swizzle_map_irq_fn())
177      *
178      * We only need one entry per interrupt in the table (not one per
179      * possible slot) seeing the interrupt-map-mask will allow the table
180      * to wrap to any number of devices.
181      */
182     for (dev = 0; dev < GPEX_NUM_IRQS; dev++) {
183         int devfn = dev * 0x8;
184 
185         for (pin = 0; pin < GPEX_NUM_IRQS; pin++) {
186             int irq_nr = PCIE_IRQ + ((pin + PCI_SLOT(devfn)) % GPEX_NUM_IRQS);
187             int i = 0;
188 
189             /* Fill PCI address cells */
190             irq_map[i] = cpu_to_be32(devfn << 8);
191             i += FDT_PCI_ADDR_CELLS;
192 
193             /* Fill PCI Interrupt cells */
194             irq_map[i] = cpu_to_be32(pin + 1);
195             i += FDT_PCI_INT_CELLS;
196 
197             /* Fill interrupt controller phandle and cells */
198             irq_map[i++] = cpu_to_be32(irqchip_phandle);
199             irq_map[i++] = cpu_to_be32(irq_nr);
200             if (s->aia_type != VIRT_AIA_TYPE_NONE) {
201                 irq_map[i++] = cpu_to_be32(0x4);
202             }
203 
204             if (!irq_map_stride) {
205                 irq_map_stride = i;
206             }
207             irq_map += irq_map_stride;
208         }
209     }
210 
211     qemu_fdt_setprop(fdt, nodename, "interrupt-map", full_irq_map,
212                      GPEX_NUM_IRQS * GPEX_NUM_IRQS *
213                      irq_map_stride * sizeof(uint32_t));
214 
215     qemu_fdt_setprop_cells(fdt, nodename, "interrupt-map-mask",
216                            0x1800, 0, 0, 0x7);
217 }
218 
219 static void create_fdt_socket_cpus(RISCVVirtState *s, int socket,
220                                    char *clust_name, uint32_t *phandle,
221                                    uint32_t *intc_phandles)
222 {
223     int cpu;
224     uint32_t cpu_phandle;
225     MachineState *ms = MACHINE(s);
226     bool is_32_bit = riscv_is_32bit(&s->soc[0]);
227     uint8_t satp_mode_max;
228 
229     for (cpu = s->soc[socket].num_harts - 1; cpu >= 0; cpu--) {
230         RISCVCPU *cpu_ptr = &s->soc[socket].harts[cpu];
231         g_autofree char *cpu_name = NULL;
232         g_autofree char *core_name = NULL;
233         g_autofree char *intc_name = NULL;
234         g_autofree char *sv_name = NULL;
235 
236         cpu_phandle = (*phandle)++;
237 
238         cpu_name = g_strdup_printf("/cpus/cpu@%d",
239             s->soc[socket].hartid_base + cpu);
240         qemu_fdt_add_subnode(ms->fdt, cpu_name);
241 
242         if (cpu_ptr->cfg.satp_mode.supported != 0) {
243             satp_mode_max = satp_mode_max_from_map(cpu_ptr->cfg.satp_mode.map);
244             sv_name = g_strdup_printf("riscv,%s",
245                                       satp_mode_str(satp_mode_max, is_32_bit));
246             qemu_fdt_setprop_string(ms->fdt, cpu_name, "mmu-type", sv_name);
247         }
248 
249         riscv_isa_write_fdt(cpu_ptr, ms->fdt, cpu_name);
250 
251         if (cpu_ptr->cfg.ext_zicbom) {
252             qemu_fdt_setprop_cell(ms->fdt, cpu_name, "riscv,cbom-block-size",
253                                   cpu_ptr->cfg.cbom_blocksize);
254         }
255 
256         if (cpu_ptr->cfg.ext_zicboz) {
257             qemu_fdt_setprop_cell(ms->fdt, cpu_name, "riscv,cboz-block-size",
258                                   cpu_ptr->cfg.cboz_blocksize);
259         }
260 
261         if (cpu_ptr->cfg.ext_zicbop) {
262             qemu_fdt_setprop_cell(ms->fdt, cpu_name, "riscv,cbop-block-size",
263                                   cpu_ptr->cfg.cbop_blocksize);
264         }
265 
266         qemu_fdt_setprop_string(ms->fdt, cpu_name, "compatible", "riscv");
267         qemu_fdt_setprop_string(ms->fdt, cpu_name, "status", "okay");
268         qemu_fdt_setprop_cell(ms->fdt, cpu_name, "reg",
269             s->soc[socket].hartid_base + cpu);
270         qemu_fdt_setprop_string(ms->fdt, cpu_name, "device_type", "cpu");
271         riscv_socket_fdt_write_id(ms, cpu_name, socket);
272         qemu_fdt_setprop_cell(ms->fdt, cpu_name, "phandle", cpu_phandle);
273 
274         intc_phandles[cpu] = (*phandle)++;
275 
276         intc_name = g_strdup_printf("%s/interrupt-controller", cpu_name);
277         qemu_fdt_add_subnode(ms->fdt, intc_name);
278         qemu_fdt_setprop_cell(ms->fdt, intc_name, "phandle",
279             intc_phandles[cpu]);
280         qemu_fdt_setprop_string(ms->fdt, intc_name, "compatible",
281             "riscv,cpu-intc");
282         qemu_fdt_setprop(ms->fdt, intc_name, "interrupt-controller", NULL, 0);
283         qemu_fdt_setprop_cell(ms->fdt, intc_name, "#interrupt-cells", 1);
284 
285         core_name = g_strdup_printf("%s/core%d", clust_name, cpu);
286         qemu_fdt_add_subnode(ms->fdt, core_name);
287         qemu_fdt_setprop_cell(ms->fdt, core_name, "cpu", cpu_phandle);
288     }
289 }
290 
291 static void create_fdt_socket_memory(RISCVVirtState *s,
292                                      const MemMapEntry *memmap, int socket)
293 {
294     g_autofree char *mem_name = NULL;
295     uint64_t addr, size;
296     MachineState *ms = MACHINE(s);
297 
298     addr = memmap[VIRT_DRAM].base + riscv_socket_mem_offset(ms, socket);
299     size = riscv_socket_mem_size(ms, socket);
300     mem_name = g_strdup_printf("/memory@%lx", (long)addr);
301     qemu_fdt_add_subnode(ms->fdt, mem_name);
302     qemu_fdt_setprop_cells(ms->fdt, mem_name, "reg",
303         addr >> 32, addr, size >> 32, size);
304     qemu_fdt_setprop_string(ms->fdt, mem_name, "device_type", "memory");
305     riscv_socket_fdt_write_id(ms, mem_name, socket);
306 }
307 
308 static void create_fdt_socket_clint(RISCVVirtState *s,
309                                     const MemMapEntry *memmap, int socket,
310                                     uint32_t *intc_phandles)
311 {
312     int cpu;
313     g_autofree char *clint_name = NULL;
314     g_autofree uint32_t *clint_cells = NULL;
315     unsigned long clint_addr;
316     MachineState *ms = MACHINE(s);
317     static const char * const clint_compat[2] = {
318         "sifive,clint0", "riscv,clint0"
319     };
320 
321     clint_cells = g_new0(uint32_t, s->soc[socket].num_harts * 4);
322 
323     for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
324         clint_cells[cpu * 4 + 0] = cpu_to_be32(intc_phandles[cpu]);
325         clint_cells[cpu * 4 + 1] = cpu_to_be32(IRQ_M_SOFT);
326         clint_cells[cpu * 4 + 2] = cpu_to_be32(intc_phandles[cpu]);
327         clint_cells[cpu * 4 + 3] = cpu_to_be32(IRQ_M_TIMER);
328     }
329 
330     clint_addr = memmap[VIRT_CLINT].base + (memmap[VIRT_CLINT].size * socket);
331     clint_name = g_strdup_printf("/soc/clint@%lx", clint_addr);
332     qemu_fdt_add_subnode(ms->fdt, clint_name);
333     qemu_fdt_setprop_string_array(ms->fdt, clint_name, "compatible",
334                                   (char **)&clint_compat,
335                                   ARRAY_SIZE(clint_compat));
336     qemu_fdt_setprop_cells(ms->fdt, clint_name, "reg",
337         0x0, clint_addr, 0x0, memmap[VIRT_CLINT].size);
338     qemu_fdt_setprop(ms->fdt, clint_name, "interrupts-extended",
339         clint_cells, s->soc[socket].num_harts * sizeof(uint32_t) * 4);
340     riscv_socket_fdt_write_id(ms, clint_name, socket);
341 }
342 
343 static void create_fdt_socket_aclint(RISCVVirtState *s,
344                                      const MemMapEntry *memmap, int socket,
345                                      uint32_t *intc_phandles)
346 {
347     int cpu;
348     char *name;
349     unsigned long addr, size;
350     uint32_t aclint_cells_size;
351     g_autofree uint32_t *aclint_mswi_cells = NULL;
352     g_autofree uint32_t *aclint_sswi_cells = NULL;
353     g_autofree uint32_t *aclint_mtimer_cells = NULL;
354     MachineState *ms = MACHINE(s);
355 
356     aclint_mswi_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
357     aclint_mtimer_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
358     aclint_sswi_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
359 
360     for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
361         aclint_mswi_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
362         aclint_mswi_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_M_SOFT);
363         aclint_mtimer_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
364         aclint_mtimer_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_M_TIMER);
365         aclint_sswi_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
366         aclint_sswi_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_S_SOFT);
367     }
368     aclint_cells_size = s->soc[socket].num_harts * sizeof(uint32_t) * 2;
369 
370     if (s->aia_type != VIRT_AIA_TYPE_APLIC_IMSIC) {
371         addr = memmap[VIRT_CLINT].base + (memmap[VIRT_CLINT].size * socket);
372         name = g_strdup_printf("/soc/mswi@%lx", addr);
373         qemu_fdt_add_subnode(ms->fdt, name);
374         qemu_fdt_setprop_string(ms->fdt, name, "compatible",
375             "riscv,aclint-mswi");
376         qemu_fdt_setprop_cells(ms->fdt, name, "reg",
377             0x0, addr, 0x0, RISCV_ACLINT_SWI_SIZE);
378         qemu_fdt_setprop(ms->fdt, name, "interrupts-extended",
379             aclint_mswi_cells, aclint_cells_size);
380         qemu_fdt_setprop(ms->fdt, name, "interrupt-controller", NULL, 0);
381         qemu_fdt_setprop_cell(ms->fdt, name, "#interrupt-cells", 0);
382         riscv_socket_fdt_write_id(ms, name, socket);
383         g_free(name);
384     }
385 
386     if (s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC) {
387         addr = memmap[VIRT_CLINT].base +
388                (RISCV_ACLINT_DEFAULT_MTIMER_SIZE * socket);
389         size = RISCV_ACLINT_DEFAULT_MTIMER_SIZE;
390     } else {
391         addr = memmap[VIRT_CLINT].base + RISCV_ACLINT_SWI_SIZE +
392             (memmap[VIRT_CLINT].size * socket);
393         size = memmap[VIRT_CLINT].size - RISCV_ACLINT_SWI_SIZE;
394     }
395     name = g_strdup_printf("/soc/mtimer@%lx", addr);
396     qemu_fdt_add_subnode(ms->fdt, name);
397     qemu_fdt_setprop_string(ms->fdt, name, "compatible",
398         "riscv,aclint-mtimer");
399     qemu_fdt_setprop_cells(ms->fdt, name, "reg",
400         0x0, addr + RISCV_ACLINT_DEFAULT_MTIME,
401         0x0, size - RISCV_ACLINT_DEFAULT_MTIME,
402         0x0, addr + RISCV_ACLINT_DEFAULT_MTIMECMP,
403         0x0, RISCV_ACLINT_DEFAULT_MTIME);
404     qemu_fdt_setprop(ms->fdt, name, "interrupts-extended",
405         aclint_mtimer_cells, aclint_cells_size);
406     riscv_socket_fdt_write_id(ms, name, socket);
407     g_free(name);
408 
409     if (s->aia_type != VIRT_AIA_TYPE_APLIC_IMSIC) {
410         addr = memmap[VIRT_ACLINT_SSWI].base +
411             (memmap[VIRT_ACLINT_SSWI].size * socket);
412         name = g_strdup_printf("/soc/sswi@%lx", addr);
413         qemu_fdt_add_subnode(ms->fdt, name);
414         qemu_fdt_setprop_string(ms->fdt, name, "compatible",
415             "riscv,aclint-sswi");
416         qemu_fdt_setprop_cells(ms->fdt, name, "reg",
417             0x0, addr, 0x0, memmap[VIRT_ACLINT_SSWI].size);
418         qemu_fdt_setprop(ms->fdt, name, "interrupts-extended",
419             aclint_sswi_cells, aclint_cells_size);
420         qemu_fdt_setprop(ms->fdt, name, "interrupt-controller", NULL, 0);
421         qemu_fdt_setprop_cell(ms->fdt, name, "#interrupt-cells", 0);
422         riscv_socket_fdt_write_id(ms, name, socket);
423         g_free(name);
424     }
425 }
426 
427 static void create_fdt_socket_plic(RISCVVirtState *s,
428                                    const MemMapEntry *memmap, int socket,
429                                    uint32_t *phandle, uint32_t *intc_phandles,
430                                    uint32_t *plic_phandles)
431 {
432     int cpu;
433     g_autofree char *plic_name = NULL;
434     g_autofree uint32_t *plic_cells;
435     unsigned long plic_addr;
436     MachineState *ms = MACHINE(s);
437     static const char * const plic_compat[2] = {
438         "sifive,plic-1.0.0", "riscv,plic0"
439     };
440 
441     plic_phandles[socket] = (*phandle)++;
442     plic_addr = memmap[VIRT_PLIC].base + (memmap[VIRT_PLIC].size * socket);
443     plic_name = g_strdup_printf("/soc/plic@%lx", plic_addr);
444     qemu_fdt_add_subnode(ms->fdt, plic_name);
445     qemu_fdt_setprop_cell(ms->fdt, plic_name,
446         "#interrupt-cells", FDT_PLIC_INT_CELLS);
447     qemu_fdt_setprop_cell(ms->fdt, plic_name,
448         "#address-cells", FDT_PLIC_ADDR_CELLS);
449     qemu_fdt_setprop_string_array(ms->fdt, plic_name, "compatible",
450                                   (char **)&plic_compat,
451                                   ARRAY_SIZE(plic_compat));
452     qemu_fdt_setprop(ms->fdt, plic_name, "interrupt-controller", NULL, 0);
453 
454     if (kvm_enabled()) {
455         plic_cells = g_new0(uint32_t, s->soc[socket].num_harts * 2);
456 
457         for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
458             plic_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
459             plic_cells[cpu * 2 + 1] = cpu_to_be32(IRQ_S_EXT);
460         }
461 
462         qemu_fdt_setprop(ms->fdt, plic_name, "interrupts-extended",
463                          plic_cells,
464                          s->soc[socket].num_harts * sizeof(uint32_t) * 2);
465    } else {
466         plic_cells = g_new0(uint32_t, s->soc[socket].num_harts * 4);
467 
468         for (cpu = 0; cpu < s->soc[socket].num_harts; cpu++) {
469             plic_cells[cpu * 4 + 0] = cpu_to_be32(intc_phandles[cpu]);
470             plic_cells[cpu * 4 + 1] = cpu_to_be32(IRQ_M_EXT);
471             plic_cells[cpu * 4 + 2] = cpu_to_be32(intc_phandles[cpu]);
472             plic_cells[cpu * 4 + 3] = cpu_to_be32(IRQ_S_EXT);
473         }
474 
475         qemu_fdt_setprop(ms->fdt, plic_name, "interrupts-extended",
476                          plic_cells,
477                          s->soc[socket].num_harts * sizeof(uint32_t) * 4);
478     }
479 
480     qemu_fdt_setprop_cells(ms->fdt, plic_name, "reg",
481         0x0, plic_addr, 0x0, memmap[VIRT_PLIC].size);
482     qemu_fdt_setprop_cell(ms->fdt, plic_name, "riscv,ndev",
483                           VIRT_IRQCHIP_NUM_SOURCES - 1);
484     riscv_socket_fdt_write_id(ms, plic_name, socket);
485     qemu_fdt_setprop_cell(ms->fdt, plic_name, "phandle",
486         plic_phandles[socket]);
487 
488     if (!socket) {
489         platform_bus_add_all_fdt_nodes(ms->fdt, plic_name,
490                                        memmap[VIRT_PLATFORM_BUS].base,
491                                        memmap[VIRT_PLATFORM_BUS].size,
492                                        VIRT_PLATFORM_BUS_IRQ);
493     }
494 }
495 
496 uint32_t imsic_num_bits(uint32_t count)
497 {
498     uint32_t ret = 0;
499 
500     while (BIT(ret) < count) {
501         ret++;
502     }
503 
504     return ret;
505 }
506 
507 static void create_fdt_one_imsic(RISCVVirtState *s, hwaddr base_addr,
508                                  uint32_t *intc_phandles, uint32_t msi_phandle,
509                                  bool m_mode, uint32_t imsic_guest_bits)
510 {
511     int cpu, socket;
512     g_autofree char *imsic_name = NULL;
513     MachineState *ms = MACHINE(s);
514     int socket_count = riscv_socket_count(ms);
515     uint32_t imsic_max_hart_per_socket, imsic_addr, imsic_size;
516     g_autofree uint32_t *imsic_cells = NULL;
517     g_autofree uint32_t *imsic_regs = NULL;
518 
519     imsic_cells = g_new0(uint32_t, ms->smp.cpus * 2);
520     imsic_regs = g_new0(uint32_t, socket_count * 4);
521 
522     for (cpu = 0; cpu < ms->smp.cpus; cpu++) {
523         imsic_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
524         imsic_cells[cpu * 2 + 1] = cpu_to_be32(m_mode ? IRQ_M_EXT : IRQ_S_EXT);
525     }
526 
527     imsic_max_hart_per_socket = 0;
528     for (socket = 0; socket < socket_count; socket++) {
529         imsic_addr = base_addr + socket * VIRT_IMSIC_GROUP_MAX_SIZE;
530         imsic_size = IMSIC_HART_SIZE(imsic_guest_bits) *
531                      s->soc[socket].num_harts;
532         imsic_regs[socket * 4 + 0] = 0;
533         imsic_regs[socket * 4 + 1] = cpu_to_be32(imsic_addr);
534         imsic_regs[socket * 4 + 2] = 0;
535         imsic_regs[socket * 4 + 3] = cpu_to_be32(imsic_size);
536         if (imsic_max_hart_per_socket < s->soc[socket].num_harts) {
537             imsic_max_hart_per_socket = s->soc[socket].num_harts;
538         }
539     }
540 
541     imsic_name = g_strdup_printf("/soc/imsics@%lx", (unsigned long)base_addr);
542     qemu_fdt_add_subnode(ms->fdt, imsic_name);
543     qemu_fdt_setprop_string(ms->fdt, imsic_name, "compatible", "riscv,imsics");
544     qemu_fdt_setprop_cell(ms->fdt, imsic_name, "#interrupt-cells",
545                           FDT_IMSIC_INT_CELLS);
546     qemu_fdt_setprop(ms->fdt, imsic_name, "interrupt-controller", NULL, 0);
547     qemu_fdt_setprop(ms->fdt, imsic_name, "msi-controller", NULL, 0);
548     qemu_fdt_setprop(ms->fdt, imsic_name, "interrupts-extended",
549                      imsic_cells, ms->smp.cpus * sizeof(uint32_t) * 2);
550     qemu_fdt_setprop(ms->fdt, imsic_name, "reg", imsic_regs,
551                      socket_count * sizeof(uint32_t) * 4);
552     qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,num-ids",
553                      VIRT_IRQCHIP_NUM_MSIS);
554 
555     if (imsic_guest_bits) {
556         qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,guest-index-bits",
557                               imsic_guest_bits);
558     }
559 
560     if (socket_count > 1) {
561         qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,hart-index-bits",
562                               imsic_num_bits(imsic_max_hart_per_socket));
563         qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,group-index-bits",
564                               imsic_num_bits(socket_count));
565         qemu_fdt_setprop_cell(ms->fdt, imsic_name, "riscv,group-index-shift",
566                               IMSIC_MMIO_GROUP_MIN_SHIFT);
567     }
568     qemu_fdt_setprop_cell(ms->fdt, imsic_name, "phandle", msi_phandle);
569 }
570 
571 static void create_fdt_imsic(RISCVVirtState *s, const MemMapEntry *memmap,
572                              uint32_t *phandle, uint32_t *intc_phandles,
573                              uint32_t *msi_m_phandle, uint32_t *msi_s_phandle)
574 {
575     *msi_m_phandle = (*phandle)++;
576     *msi_s_phandle = (*phandle)++;
577 
578     if (!kvm_enabled()) {
579         /* M-level IMSIC node */
580         create_fdt_one_imsic(s, memmap[VIRT_IMSIC_M].base, intc_phandles,
581                              *msi_m_phandle, true, 0);
582     }
583 
584     /* S-level IMSIC node */
585     create_fdt_one_imsic(s, memmap[VIRT_IMSIC_S].base, intc_phandles,
586                          *msi_s_phandle, false,
587                          imsic_num_bits(s->aia_guests + 1));
588 
589 }
590 
591 static void create_fdt_one_aplic(RISCVVirtState *s, int socket,
592                                  unsigned long aplic_addr, uint32_t aplic_size,
593                                  uint32_t msi_phandle,
594                                  uint32_t *intc_phandles,
595                                  uint32_t aplic_phandle,
596                                  uint32_t aplic_child_phandle,
597                                  bool m_mode, int num_harts)
598 {
599     int cpu;
600     g_autofree char *aplic_name = NULL;
601     g_autofree uint32_t *aplic_cells = g_new0(uint32_t, num_harts * 2);
602     MachineState *ms = MACHINE(s);
603 
604     for (cpu = 0; cpu < num_harts; cpu++) {
605         aplic_cells[cpu * 2 + 0] = cpu_to_be32(intc_phandles[cpu]);
606         aplic_cells[cpu * 2 + 1] = cpu_to_be32(m_mode ? IRQ_M_EXT : IRQ_S_EXT);
607     }
608 
609     aplic_name = g_strdup_printf("/soc/aplic@%lx", aplic_addr);
610     qemu_fdt_add_subnode(ms->fdt, aplic_name);
611     qemu_fdt_setprop_string(ms->fdt, aplic_name, "compatible", "riscv,aplic");
612     qemu_fdt_setprop_cell(ms->fdt, aplic_name,
613                           "#interrupt-cells", FDT_APLIC_INT_CELLS);
614     qemu_fdt_setprop(ms->fdt, aplic_name, "interrupt-controller", NULL, 0);
615 
616     if (s->aia_type == VIRT_AIA_TYPE_APLIC) {
617         qemu_fdt_setprop(ms->fdt, aplic_name, "interrupts-extended",
618                          aplic_cells, num_harts * sizeof(uint32_t) * 2);
619     } else {
620         qemu_fdt_setprop_cell(ms->fdt, aplic_name, "msi-parent", msi_phandle);
621     }
622 
623     qemu_fdt_setprop_cells(ms->fdt, aplic_name, "reg",
624                            0x0, aplic_addr, 0x0, aplic_size);
625     qemu_fdt_setprop_cell(ms->fdt, aplic_name, "riscv,num-sources",
626                           VIRT_IRQCHIP_NUM_SOURCES);
627 
628     if (aplic_child_phandle) {
629         qemu_fdt_setprop_cell(ms->fdt, aplic_name, "riscv,children",
630                               aplic_child_phandle);
631         qemu_fdt_setprop_cells(ms->fdt, aplic_name, "riscv,delegate",
632                                aplic_child_phandle, 0x1,
633                                VIRT_IRQCHIP_NUM_SOURCES);
634     }
635 
636     riscv_socket_fdt_write_id(ms, aplic_name, socket);
637     qemu_fdt_setprop_cell(ms->fdt, aplic_name, "phandle", aplic_phandle);
638 }
639 
640 static void create_fdt_socket_aplic(RISCVVirtState *s,
641                                     const MemMapEntry *memmap, int socket,
642                                     uint32_t msi_m_phandle,
643                                     uint32_t msi_s_phandle,
644                                     uint32_t *phandle,
645                                     uint32_t *intc_phandles,
646                                     uint32_t *aplic_phandles,
647                                     int num_harts)
648 {
649     g_autofree char *aplic_name = NULL;
650     unsigned long aplic_addr;
651     MachineState *ms = MACHINE(s);
652     uint32_t aplic_m_phandle, aplic_s_phandle;
653 
654     aplic_m_phandle = (*phandle)++;
655     aplic_s_phandle = (*phandle)++;
656 
657     if (!kvm_enabled()) {
658         /* M-level APLIC node */
659         aplic_addr = memmap[VIRT_APLIC_M].base +
660                      (memmap[VIRT_APLIC_M].size * socket);
661         create_fdt_one_aplic(s, socket, aplic_addr, memmap[VIRT_APLIC_M].size,
662                              msi_m_phandle, intc_phandles,
663                              aplic_m_phandle, aplic_s_phandle,
664                              true, num_harts);
665     }
666 
667     /* S-level APLIC node */
668     aplic_addr = memmap[VIRT_APLIC_S].base +
669                  (memmap[VIRT_APLIC_S].size * socket);
670     create_fdt_one_aplic(s, socket, aplic_addr, memmap[VIRT_APLIC_S].size,
671                          msi_s_phandle, intc_phandles,
672                          aplic_s_phandle, 0,
673                          false, num_harts);
674 
675     aplic_name = g_strdup_printf("/soc/aplic@%lx", aplic_addr);
676 
677     if (!socket) {
678         platform_bus_add_all_fdt_nodes(ms->fdt, aplic_name,
679                                        memmap[VIRT_PLATFORM_BUS].base,
680                                        memmap[VIRT_PLATFORM_BUS].size,
681                                        VIRT_PLATFORM_BUS_IRQ);
682     }
683 
684     aplic_phandles[socket] = aplic_s_phandle;
685 }
686 
687 static void create_fdt_pmu(RISCVVirtState *s)
688 {
689     g_autofree char *pmu_name = g_strdup_printf("/pmu");
690     MachineState *ms = MACHINE(s);
691     RISCVCPU hart = s->soc[0].harts[0];
692 
693     qemu_fdt_add_subnode(ms->fdt, pmu_name);
694     qemu_fdt_setprop_string(ms->fdt, pmu_name, "compatible", "riscv,pmu");
695     riscv_pmu_generate_fdt_node(ms->fdt, hart.pmu_avail_ctrs, pmu_name);
696 }
697 
698 static void create_fdt_sockets(RISCVVirtState *s, const MemMapEntry *memmap,
699                                uint32_t *phandle,
700                                uint32_t *irq_mmio_phandle,
701                                uint32_t *irq_pcie_phandle,
702                                uint32_t *irq_virtio_phandle,
703                                uint32_t *msi_pcie_phandle)
704 {
705     int socket, phandle_pos;
706     MachineState *ms = MACHINE(s);
707     uint32_t msi_m_phandle = 0, msi_s_phandle = 0;
708     uint32_t xplic_phandles[MAX_NODES];
709     g_autofree uint32_t *intc_phandles = NULL;
710     int socket_count = riscv_socket_count(ms);
711 
712     qemu_fdt_add_subnode(ms->fdt, "/cpus");
713     qemu_fdt_setprop_cell(ms->fdt, "/cpus", "timebase-frequency",
714                           kvm_enabled() ?
715                           kvm_riscv_get_timebase_frequency(first_cpu) :
716                           RISCV_ACLINT_DEFAULT_TIMEBASE_FREQ);
717     qemu_fdt_setprop_cell(ms->fdt, "/cpus", "#size-cells", 0x0);
718     qemu_fdt_setprop_cell(ms->fdt, "/cpus", "#address-cells", 0x1);
719     qemu_fdt_add_subnode(ms->fdt, "/cpus/cpu-map");
720 
721     intc_phandles = g_new0(uint32_t, ms->smp.cpus);
722 
723     phandle_pos = ms->smp.cpus;
724     for (socket = (socket_count - 1); socket >= 0; socket--) {
725         g_autofree char *clust_name = NULL;
726         phandle_pos -= s->soc[socket].num_harts;
727 
728         clust_name = g_strdup_printf("/cpus/cpu-map/cluster%d", socket);
729         qemu_fdt_add_subnode(ms->fdt, clust_name);
730 
731         create_fdt_socket_cpus(s, socket, clust_name, phandle,
732                                &intc_phandles[phandle_pos]);
733 
734         create_fdt_socket_memory(s, memmap, socket);
735 
736         if (virt_aclint_allowed() && s->have_aclint) {
737             create_fdt_socket_aclint(s, memmap, socket,
738                                      &intc_phandles[phandle_pos]);
739         } else if (tcg_enabled()) {
740             create_fdt_socket_clint(s, memmap, socket,
741                                     &intc_phandles[phandle_pos]);
742         }
743     }
744 
745     if (s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC) {
746         create_fdt_imsic(s, memmap, phandle, intc_phandles,
747             &msi_m_phandle, &msi_s_phandle);
748         *msi_pcie_phandle = msi_s_phandle;
749     }
750 
751     /* KVM AIA only has one APLIC instance */
752     if (kvm_enabled() && virt_use_kvm_aia(s)) {
753         create_fdt_socket_aplic(s, memmap, 0,
754                                 msi_m_phandle, msi_s_phandle, phandle,
755                                 &intc_phandles[0], xplic_phandles,
756                                 ms->smp.cpus);
757     } else {
758         phandle_pos = ms->smp.cpus;
759         for (socket = (socket_count - 1); socket >= 0; socket--) {
760             phandle_pos -= s->soc[socket].num_harts;
761 
762             if (s->aia_type == VIRT_AIA_TYPE_NONE) {
763                 create_fdt_socket_plic(s, memmap, socket, phandle,
764                                        &intc_phandles[phandle_pos],
765                                        xplic_phandles);
766             } else {
767                 create_fdt_socket_aplic(s, memmap, socket,
768                                         msi_m_phandle, msi_s_phandle, phandle,
769                                         &intc_phandles[phandle_pos],
770                                         xplic_phandles,
771                                         s->soc[socket].num_harts);
772             }
773         }
774     }
775 
776     if (kvm_enabled() && virt_use_kvm_aia(s)) {
777         *irq_mmio_phandle = xplic_phandles[0];
778         *irq_virtio_phandle = xplic_phandles[0];
779         *irq_pcie_phandle = xplic_phandles[0];
780     } else {
781         for (socket = 0; socket < socket_count; socket++) {
782             if (socket == 0) {
783                 *irq_mmio_phandle = xplic_phandles[socket];
784                 *irq_virtio_phandle = xplic_phandles[socket];
785                 *irq_pcie_phandle = xplic_phandles[socket];
786             }
787             if (socket == 1) {
788                 *irq_virtio_phandle = xplic_phandles[socket];
789                 *irq_pcie_phandle = xplic_phandles[socket];
790             }
791             if (socket == 2) {
792                 *irq_pcie_phandle = xplic_phandles[socket];
793             }
794         }
795     }
796 
797     riscv_socket_fdt_write_distance_matrix(ms);
798 }
799 
800 static void create_fdt_virtio(RISCVVirtState *s, const MemMapEntry *memmap,
801                               uint32_t irq_virtio_phandle)
802 {
803     int i;
804     MachineState *ms = MACHINE(s);
805 
806     for (i = 0; i < VIRTIO_COUNT; i++) {
807         g_autofree char *name =  g_strdup_printf("/soc/virtio_mmio@%lx",
808             (long)(memmap[VIRT_VIRTIO].base + i * memmap[VIRT_VIRTIO].size));
809 
810         qemu_fdt_add_subnode(ms->fdt, name);
811         qemu_fdt_setprop_string(ms->fdt, name, "compatible", "virtio,mmio");
812         qemu_fdt_setprop_cells(ms->fdt, name, "reg",
813             0x0, memmap[VIRT_VIRTIO].base + i * memmap[VIRT_VIRTIO].size,
814             0x0, memmap[VIRT_VIRTIO].size);
815         qemu_fdt_setprop_cell(ms->fdt, name, "interrupt-parent",
816             irq_virtio_phandle);
817         if (s->aia_type == VIRT_AIA_TYPE_NONE) {
818             qemu_fdt_setprop_cell(ms->fdt, name, "interrupts",
819                                   VIRTIO_IRQ + i);
820         } else {
821             qemu_fdt_setprop_cells(ms->fdt, name, "interrupts",
822                                    VIRTIO_IRQ + i, 0x4);
823         }
824     }
825 }
826 
827 static void create_fdt_pcie(RISCVVirtState *s, const MemMapEntry *memmap,
828                             uint32_t irq_pcie_phandle,
829                             uint32_t msi_pcie_phandle)
830 {
831     g_autofree char *name = NULL;
832     MachineState *ms = MACHINE(s);
833 
834     name = g_strdup_printf("/soc/pci@%lx",
835         (long) memmap[VIRT_PCIE_ECAM].base);
836     qemu_fdt_setprop_cell(ms->fdt, name, "#address-cells",
837         FDT_PCI_ADDR_CELLS);
838     qemu_fdt_setprop_cell(ms->fdt, name, "#interrupt-cells",
839         FDT_PCI_INT_CELLS);
840     qemu_fdt_setprop_cell(ms->fdt, name, "#size-cells", 0x2);
841     qemu_fdt_setprop_string(ms->fdt, name, "compatible",
842         "pci-host-ecam-generic");
843     qemu_fdt_setprop_string(ms->fdt, name, "device_type", "pci");
844     qemu_fdt_setprop_cell(ms->fdt, name, "linux,pci-domain", 0);
845     qemu_fdt_setprop_cells(ms->fdt, name, "bus-range", 0,
846         memmap[VIRT_PCIE_ECAM].size / PCIE_MMCFG_SIZE_MIN - 1);
847     qemu_fdt_setprop(ms->fdt, name, "dma-coherent", NULL, 0);
848     if (s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC) {
849         qemu_fdt_setprop_cell(ms->fdt, name, "msi-parent", msi_pcie_phandle);
850     }
851     qemu_fdt_setprop_cells(ms->fdt, name, "reg", 0,
852         memmap[VIRT_PCIE_ECAM].base, 0, memmap[VIRT_PCIE_ECAM].size);
853     qemu_fdt_setprop_sized_cells(ms->fdt, name, "ranges",
854         1, FDT_PCI_RANGE_IOPORT, 2, 0,
855         2, memmap[VIRT_PCIE_PIO].base, 2, memmap[VIRT_PCIE_PIO].size,
856         1, FDT_PCI_RANGE_MMIO,
857         2, memmap[VIRT_PCIE_MMIO].base,
858         2, memmap[VIRT_PCIE_MMIO].base, 2, memmap[VIRT_PCIE_MMIO].size,
859         1, FDT_PCI_RANGE_MMIO_64BIT,
860         2, virt_high_pcie_memmap.base,
861         2, virt_high_pcie_memmap.base, 2, virt_high_pcie_memmap.size);
862 
863     create_pcie_irq_map(s, ms->fdt, name, irq_pcie_phandle);
864 }
865 
866 static void create_fdt_reset(RISCVVirtState *s, const MemMapEntry *memmap,
867                              uint32_t *phandle)
868 {
869     char *name;
870     uint32_t test_phandle;
871     MachineState *ms = MACHINE(s);
872 
873     test_phandle = (*phandle)++;
874     name = g_strdup_printf("/soc/test@%lx",
875         (long)memmap[VIRT_TEST].base);
876     qemu_fdt_add_subnode(ms->fdt, name);
877     {
878         static const char * const compat[3] = {
879             "sifive,test1", "sifive,test0", "syscon"
880         };
881         qemu_fdt_setprop_string_array(ms->fdt, name, "compatible",
882                                       (char **)&compat, ARRAY_SIZE(compat));
883     }
884     qemu_fdt_setprop_cells(ms->fdt, name, "reg",
885         0x0, memmap[VIRT_TEST].base, 0x0, memmap[VIRT_TEST].size);
886     qemu_fdt_setprop_cell(ms->fdt, name, "phandle", test_phandle);
887     test_phandle = qemu_fdt_get_phandle(ms->fdt, name);
888     g_free(name);
889 
890     name = g_strdup_printf("/reboot");
891     qemu_fdt_add_subnode(ms->fdt, name);
892     qemu_fdt_setprop_string(ms->fdt, name, "compatible", "syscon-reboot");
893     qemu_fdt_setprop_cell(ms->fdt, name, "regmap", test_phandle);
894     qemu_fdt_setprop_cell(ms->fdt, name, "offset", 0x0);
895     qemu_fdt_setprop_cell(ms->fdt, name, "value", FINISHER_RESET);
896     g_free(name);
897 
898     name = g_strdup_printf("/poweroff");
899     qemu_fdt_add_subnode(ms->fdt, name);
900     qemu_fdt_setprop_string(ms->fdt, name, "compatible", "syscon-poweroff");
901     qemu_fdt_setprop_cell(ms->fdt, name, "regmap", test_phandle);
902     qemu_fdt_setprop_cell(ms->fdt, name, "offset", 0x0);
903     qemu_fdt_setprop_cell(ms->fdt, name, "value", FINISHER_PASS);
904     g_free(name);
905 }
906 
907 static void create_fdt_uart(RISCVVirtState *s, const MemMapEntry *memmap,
908                             uint32_t irq_mmio_phandle)
909 {
910     g_autofree char *name = NULL;
911     MachineState *ms = MACHINE(s);
912 
913     name = g_strdup_printf("/soc/serial@%lx", (long)memmap[VIRT_UART0].base);
914     qemu_fdt_add_subnode(ms->fdt, name);
915     qemu_fdt_setprop_string(ms->fdt, name, "compatible", "ns16550a");
916     qemu_fdt_setprop_cells(ms->fdt, name, "reg",
917         0x0, memmap[VIRT_UART0].base,
918         0x0, memmap[VIRT_UART0].size);
919     qemu_fdt_setprop_cell(ms->fdt, name, "clock-frequency", 3686400);
920     qemu_fdt_setprop_cell(ms->fdt, name, "interrupt-parent", irq_mmio_phandle);
921     if (s->aia_type == VIRT_AIA_TYPE_NONE) {
922         qemu_fdt_setprop_cell(ms->fdt, name, "interrupts", UART0_IRQ);
923     } else {
924         qemu_fdt_setprop_cells(ms->fdt, name, "interrupts", UART0_IRQ, 0x4);
925     }
926 
927     qemu_fdt_setprop_string(ms->fdt, "/chosen", "stdout-path", name);
928 }
929 
930 static void create_fdt_rtc(RISCVVirtState *s, const MemMapEntry *memmap,
931                            uint32_t irq_mmio_phandle)
932 {
933     g_autofree char *name = NULL;
934     MachineState *ms = MACHINE(s);
935 
936     name = g_strdup_printf("/soc/rtc@%lx", (long)memmap[VIRT_RTC].base);
937     qemu_fdt_add_subnode(ms->fdt, name);
938     qemu_fdt_setprop_string(ms->fdt, name, "compatible",
939         "google,goldfish-rtc");
940     qemu_fdt_setprop_cells(ms->fdt, name, "reg",
941         0x0, memmap[VIRT_RTC].base, 0x0, memmap[VIRT_RTC].size);
942     qemu_fdt_setprop_cell(ms->fdt, name, "interrupt-parent",
943         irq_mmio_phandle);
944     if (s->aia_type == VIRT_AIA_TYPE_NONE) {
945         qemu_fdt_setprop_cell(ms->fdt, name, "interrupts", RTC_IRQ);
946     } else {
947         qemu_fdt_setprop_cells(ms->fdt, name, "interrupts", RTC_IRQ, 0x4);
948     }
949 }
950 
951 static void create_fdt_flash(RISCVVirtState *s, const MemMapEntry *memmap)
952 {
953     MachineState *ms = MACHINE(s);
954     hwaddr flashsize = virt_memmap[VIRT_FLASH].size / 2;
955     hwaddr flashbase = virt_memmap[VIRT_FLASH].base;
956     g_autofree char *name = g_strdup_printf("/flash@%" PRIx64, flashbase);
957 
958     qemu_fdt_add_subnode(ms->fdt, name);
959     qemu_fdt_setprop_string(ms->fdt, name, "compatible", "cfi-flash");
960     qemu_fdt_setprop_sized_cells(ms->fdt, name, "reg",
961                                  2, flashbase, 2, flashsize,
962                                  2, flashbase + flashsize, 2, flashsize);
963     qemu_fdt_setprop_cell(ms->fdt, name, "bank-width", 4);
964 }
965 
966 static void create_fdt_fw_cfg(RISCVVirtState *s, const MemMapEntry *memmap)
967 {
968     MachineState *ms = MACHINE(s);
969     hwaddr base = memmap[VIRT_FW_CFG].base;
970     hwaddr size = memmap[VIRT_FW_CFG].size;
971     g_autofree char *nodename = g_strdup_printf("/fw-cfg@%" PRIx64, base);
972 
973     qemu_fdt_add_subnode(ms->fdt, nodename);
974     qemu_fdt_setprop_string(ms->fdt, nodename,
975                             "compatible", "qemu,fw-cfg-mmio");
976     qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
977                                  2, base, 2, size);
978     qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
979 }
980 
981 static void create_fdt_virtio_iommu(RISCVVirtState *s, uint16_t bdf)
982 {
983     const char compat[] = "virtio,pci-iommu\0pci1af4,1057";
984     void *fdt = MACHINE(s)->fdt;
985     uint32_t iommu_phandle;
986     g_autofree char *iommu_node = NULL;
987     g_autofree char *pci_node = NULL;
988 
989     pci_node = g_strdup_printf("/soc/pci@%lx",
990                                (long) virt_memmap[VIRT_PCIE_ECAM].base);
991     iommu_node = g_strdup_printf("%s/virtio_iommu@%x,%x", pci_node,
992                                  PCI_SLOT(bdf), PCI_FUNC(bdf));
993     iommu_phandle = qemu_fdt_alloc_phandle(fdt);
994 
995     qemu_fdt_add_subnode(fdt, iommu_node);
996 
997     qemu_fdt_setprop(fdt, iommu_node, "compatible", compat, sizeof(compat));
998     qemu_fdt_setprop_sized_cells(fdt, iommu_node, "reg",
999                                  1, bdf << 8, 1, 0, 1, 0,
1000                                  1, 0, 1, 0);
1001     qemu_fdt_setprop_cell(fdt, iommu_node, "#iommu-cells", 1);
1002     qemu_fdt_setprop_cell(fdt, iommu_node, "phandle", iommu_phandle);
1003 
1004     qemu_fdt_setprop_cells(fdt, pci_node, "iommu-map",
1005                            0, iommu_phandle, 0, bdf,
1006                            bdf + 1, iommu_phandle, bdf + 1, 0xffff - bdf);
1007 }
1008 
1009 static void finalize_fdt(RISCVVirtState *s)
1010 {
1011     uint32_t phandle = 1, irq_mmio_phandle = 1, msi_pcie_phandle = 1;
1012     uint32_t irq_pcie_phandle = 1, irq_virtio_phandle = 1;
1013 
1014     create_fdt_sockets(s, virt_memmap, &phandle, &irq_mmio_phandle,
1015                        &irq_pcie_phandle, &irq_virtio_phandle,
1016                        &msi_pcie_phandle);
1017 
1018     create_fdt_virtio(s, virt_memmap, irq_virtio_phandle);
1019 
1020     create_fdt_pcie(s, virt_memmap, irq_pcie_phandle, msi_pcie_phandle);
1021 
1022     create_fdt_reset(s, virt_memmap, &phandle);
1023 
1024     create_fdt_uart(s, virt_memmap, irq_mmio_phandle);
1025 
1026     create_fdt_rtc(s, virt_memmap, irq_mmio_phandle);
1027 }
1028 
1029 static void create_fdt(RISCVVirtState *s, const MemMapEntry *memmap)
1030 {
1031     MachineState *ms = MACHINE(s);
1032     uint8_t rng_seed[32];
1033     g_autofree char *name = NULL;
1034 
1035     ms->fdt = create_device_tree(&s->fdt_size);
1036     if (!ms->fdt) {
1037         error_report("create_device_tree() failed");
1038         exit(1);
1039     }
1040 
1041     qemu_fdt_setprop_string(ms->fdt, "/", "model", "riscv-virtio,qemu");
1042     qemu_fdt_setprop_string(ms->fdt, "/", "compatible", "riscv-virtio");
1043     qemu_fdt_setprop_cell(ms->fdt, "/", "#size-cells", 0x2);
1044     qemu_fdt_setprop_cell(ms->fdt, "/", "#address-cells", 0x2);
1045 
1046     qemu_fdt_add_subnode(ms->fdt, "/soc");
1047     qemu_fdt_setprop(ms->fdt, "/soc", "ranges", NULL, 0);
1048     qemu_fdt_setprop_string(ms->fdt, "/soc", "compatible", "simple-bus");
1049     qemu_fdt_setprop_cell(ms->fdt, "/soc", "#size-cells", 0x2);
1050     qemu_fdt_setprop_cell(ms->fdt, "/soc", "#address-cells", 0x2);
1051 
1052     /*
1053      * The "/soc/pci@..." node is needed for PCIE hotplugs
1054      * that might happen before finalize_fdt().
1055      */
1056     name = g_strdup_printf("/soc/pci@%lx", (long) memmap[VIRT_PCIE_ECAM].base);
1057     qemu_fdt_add_subnode(ms->fdt, name);
1058 
1059     qemu_fdt_add_subnode(ms->fdt, "/chosen");
1060 
1061     /* Pass seed to RNG */
1062     qemu_guest_getrandom_nofail(rng_seed, sizeof(rng_seed));
1063     qemu_fdt_setprop(ms->fdt, "/chosen", "rng-seed",
1064                      rng_seed, sizeof(rng_seed));
1065 
1066     create_fdt_flash(s, memmap);
1067     create_fdt_fw_cfg(s, memmap);
1068     create_fdt_pmu(s);
1069 }
1070 
1071 static inline DeviceState *gpex_pcie_init(MemoryRegion *sys_mem,
1072                                           DeviceState *irqchip,
1073                                           RISCVVirtState *s)
1074 {
1075     DeviceState *dev;
1076     MemoryRegion *ecam_alias, *ecam_reg;
1077     MemoryRegion *mmio_alias, *high_mmio_alias, *mmio_reg;
1078     hwaddr ecam_base = s->memmap[VIRT_PCIE_ECAM].base;
1079     hwaddr ecam_size = s->memmap[VIRT_PCIE_ECAM].size;
1080     hwaddr mmio_base = s->memmap[VIRT_PCIE_MMIO].base;
1081     hwaddr mmio_size = s->memmap[VIRT_PCIE_MMIO].size;
1082     hwaddr high_mmio_base = virt_high_pcie_memmap.base;
1083     hwaddr high_mmio_size = virt_high_pcie_memmap.size;
1084     hwaddr pio_base = s->memmap[VIRT_PCIE_PIO].base;
1085     hwaddr pio_size = s->memmap[VIRT_PCIE_PIO].size;
1086     qemu_irq irq;
1087     int i;
1088 
1089     dev = qdev_new(TYPE_GPEX_HOST);
1090 
1091     /* Set GPEX object properties for the virt machine */
1092     object_property_set_uint(OBJECT(GPEX_HOST(dev)), PCI_HOST_ECAM_BASE,
1093                             ecam_base, NULL);
1094     object_property_set_int(OBJECT(GPEX_HOST(dev)), PCI_HOST_ECAM_SIZE,
1095                             ecam_size, NULL);
1096     object_property_set_uint(OBJECT(GPEX_HOST(dev)),
1097                              PCI_HOST_BELOW_4G_MMIO_BASE,
1098                              mmio_base, NULL);
1099     object_property_set_int(OBJECT(GPEX_HOST(dev)), PCI_HOST_BELOW_4G_MMIO_SIZE,
1100                             mmio_size, NULL);
1101     object_property_set_uint(OBJECT(GPEX_HOST(dev)),
1102                              PCI_HOST_ABOVE_4G_MMIO_BASE,
1103                              high_mmio_base, NULL);
1104     object_property_set_int(OBJECT(GPEX_HOST(dev)), PCI_HOST_ABOVE_4G_MMIO_SIZE,
1105                             high_mmio_size, NULL);
1106     object_property_set_uint(OBJECT(GPEX_HOST(dev)), PCI_HOST_PIO_BASE,
1107                             pio_base, NULL);
1108     object_property_set_int(OBJECT(GPEX_HOST(dev)), PCI_HOST_PIO_SIZE,
1109                             pio_size, NULL);
1110 
1111     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1112 
1113     ecam_alias = g_new0(MemoryRegion, 1);
1114     ecam_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0);
1115     memory_region_init_alias(ecam_alias, OBJECT(dev), "pcie-ecam",
1116                              ecam_reg, 0, ecam_size);
1117     memory_region_add_subregion(get_system_memory(), ecam_base, ecam_alias);
1118 
1119     mmio_alias = g_new0(MemoryRegion, 1);
1120     mmio_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 1);
1121     memory_region_init_alias(mmio_alias, OBJECT(dev), "pcie-mmio",
1122                              mmio_reg, mmio_base, mmio_size);
1123     memory_region_add_subregion(get_system_memory(), mmio_base, mmio_alias);
1124 
1125     /* Map high MMIO space */
1126     high_mmio_alias = g_new0(MemoryRegion, 1);
1127     memory_region_init_alias(high_mmio_alias, OBJECT(dev), "pcie-mmio-high",
1128                              mmio_reg, high_mmio_base, high_mmio_size);
1129     memory_region_add_subregion(get_system_memory(), high_mmio_base,
1130                                 high_mmio_alias);
1131 
1132     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 2, pio_base);
1133 
1134     for (i = 0; i < GPEX_NUM_IRQS; i++) {
1135         irq = qdev_get_gpio_in(irqchip, PCIE_IRQ + i);
1136 
1137         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i, irq);
1138         gpex_set_irq_num(GPEX_HOST(dev), i, PCIE_IRQ + i);
1139     }
1140 
1141     GPEX_HOST(dev)->gpex_cfg.bus = PCI_HOST_BRIDGE(GPEX_HOST(dev))->bus;
1142     return dev;
1143 }
1144 
1145 static FWCfgState *create_fw_cfg(const MachineState *ms)
1146 {
1147     hwaddr base = virt_memmap[VIRT_FW_CFG].base;
1148     FWCfgState *fw_cfg;
1149 
1150     fw_cfg = fw_cfg_init_mem_wide(base + 8, base, 8, base + 16,
1151                                   &address_space_memory);
1152     fw_cfg_add_i16(fw_cfg, FW_CFG_NB_CPUS, (uint16_t)ms->smp.cpus);
1153 
1154     return fw_cfg;
1155 }
1156 
1157 static DeviceState *virt_create_plic(const MemMapEntry *memmap, int socket,
1158                                      int base_hartid, int hart_count)
1159 {
1160     DeviceState *ret;
1161     g_autofree char *plic_hart_config = NULL;
1162 
1163     /* Per-socket PLIC hart topology configuration string */
1164     plic_hart_config = riscv_plic_hart_config_string(hart_count);
1165 
1166     /* Per-socket PLIC */
1167     ret = sifive_plic_create(
1168             memmap[VIRT_PLIC].base + socket * memmap[VIRT_PLIC].size,
1169             plic_hart_config, hart_count, base_hartid,
1170             VIRT_IRQCHIP_NUM_SOURCES,
1171             ((1U << VIRT_IRQCHIP_NUM_PRIO_BITS) - 1),
1172             VIRT_PLIC_PRIORITY_BASE,
1173             VIRT_PLIC_PENDING_BASE,
1174             VIRT_PLIC_ENABLE_BASE,
1175             VIRT_PLIC_ENABLE_STRIDE,
1176             VIRT_PLIC_CONTEXT_BASE,
1177             VIRT_PLIC_CONTEXT_STRIDE,
1178             memmap[VIRT_PLIC].size);
1179 
1180     return ret;
1181 }
1182 
1183 static DeviceState *virt_create_aia(RISCVVirtAIAType aia_type, int aia_guests,
1184                                     const MemMapEntry *memmap, int socket,
1185                                     int base_hartid, int hart_count)
1186 {
1187     int i;
1188     hwaddr addr;
1189     uint32_t guest_bits;
1190     DeviceState *aplic_s = NULL;
1191     DeviceState *aplic_m = NULL;
1192     bool msimode = aia_type == VIRT_AIA_TYPE_APLIC_IMSIC;
1193 
1194     if (msimode) {
1195         if (!kvm_enabled()) {
1196             /* Per-socket M-level IMSICs */
1197             addr = memmap[VIRT_IMSIC_M].base +
1198                    socket * VIRT_IMSIC_GROUP_MAX_SIZE;
1199             for (i = 0; i < hart_count; i++) {
1200                 riscv_imsic_create(addr + i * IMSIC_HART_SIZE(0),
1201                                    base_hartid + i, true, 1,
1202                                    VIRT_IRQCHIP_NUM_MSIS);
1203             }
1204         }
1205 
1206         /* Per-socket S-level IMSICs */
1207         guest_bits = imsic_num_bits(aia_guests + 1);
1208         addr = memmap[VIRT_IMSIC_S].base + socket * VIRT_IMSIC_GROUP_MAX_SIZE;
1209         for (i = 0; i < hart_count; i++) {
1210             riscv_imsic_create(addr + i * IMSIC_HART_SIZE(guest_bits),
1211                                base_hartid + i, false, 1 + aia_guests,
1212                                VIRT_IRQCHIP_NUM_MSIS);
1213         }
1214     }
1215 
1216     if (!kvm_enabled()) {
1217         /* Per-socket M-level APLIC */
1218         aplic_m = riscv_aplic_create(memmap[VIRT_APLIC_M].base +
1219                                      socket * memmap[VIRT_APLIC_M].size,
1220                                      memmap[VIRT_APLIC_M].size,
1221                                      (msimode) ? 0 : base_hartid,
1222                                      (msimode) ? 0 : hart_count,
1223                                      VIRT_IRQCHIP_NUM_SOURCES,
1224                                      VIRT_IRQCHIP_NUM_PRIO_BITS,
1225                                      msimode, true, NULL);
1226     }
1227 
1228     /* Per-socket S-level APLIC */
1229     aplic_s = riscv_aplic_create(memmap[VIRT_APLIC_S].base +
1230                                  socket * memmap[VIRT_APLIC_S].size,
1231                                  memmap[VIRT_APLIC_S].size,
1232                                  (msimode) ? 0 : base_hartid,
1233                                  (msimode) ? 0 : hart_count,
1234                                  VIRT_IRQCHIP_NUM_SOURCES,
1235                                  VIRT_IRQCHIP_NUM_PRIO_BITS,
1236                                  msimode, false, aplic_m);
1237 
1238     return kvm_enabled() ? aplic_s : aplic_m;
1239 }
1240 
1241 static void create_platform_bus(RISCVVirtState *s, DeviceState *irqchip)
1242 {
1243     DeviceState *dev;
1244     SysBusDevice *sysbus;
1245     const MemMapEntry *memmap = virt_memmap;
1246     int i;
1247     MemoryRegion *sysmem = get_system_memory();
1248 
1249     dev = qdev_new(TYPE_PLATFORM_BUS_DEVICE);
1250     dev->id = g_strdup(TYPE_PLATFORM_BUS_DEVICE);
1251     qdev_prop_set_uint32(dev, "num_irqs", VIRT_PLATFORM_BUS_NUM_IRQS);
1252     qdev_prop_set_uint32(dev, "mmio_size", memmap[VIRT_PLATFORM_BUS].size);
1253     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1254     s->platform_bus_dev = dev;
1255 
1256     sysbus = SYS_BUS_DEVICE(dev);
1257     for (i = 0; i < VIRT_PLATFORM_BUS_NUM_IRQS; i++) {
1258         int irq = VIRT_PLATFORM_BUS_IRQ + i;
1259         sysbus_connect_irq(sysbus, i, qdev_get_gpio_in(irqchip, irq));
1260     }
1261 
1262     memory_region_add_subregion(sysmem,
1263                                 memmap[VIRT_PLATFORM_BUS].base,
1264                                 sysbus_mmio_get_region(sysbus, 0));
1265 }
1266 
1267 static void virt_build_smbios(RISCVVirtState *s)
1268 {
1269     MachineClass *mc = MACHINE_GET_CLASS(s);
1270     MachineState *ms = MACHINE(s);
1271     uint8_t *smbios_tables, *smbios_anchor;
1272     size_t smbios_tables_len, smbios_anchor_len;
1273     struct smbios_phys_mem_area mem_array;
1274     const char *product = "QEMU Virtual Machine";
1275 
1276     if (kvm_enabled()) {
1277         product = "KVM Virtual Machine";
1278     }
1279 
1280     smbios_set_defaults("QEMU", product, mc->name);
1281 
1282     if (riscv_is_32bit(&s->soc[0])) {
1283         smbios_set_default_processor_family(0x200);
1284     } else {
1285         smbios_set_default_processor_family(0x201);
1286     }
1287 
1288     /* build the array of physical mem area from base_memmap */
1289     mem_array.address = s->memmap[VIRT_DRAM].base;
1290     mem_array.length = ms->ram_size;
1291 
1292     smbios_get_tables(ms, SMBIOS_ENTRY_POINT_TYPE_64,
1293                       &mem_array, 1,
1294                       &smbios_tables, &smbios_tables_len,
1295                       &smbios_anchor, &smbios_anchor_len,
1296                       &error_fatal);
1297 
1298     if (smbios_anchor) {
1299         fw_cfg_add_file(s->fw_cfg, "etc/smbios/smbios-tables",
1300                         smbios_tables, smbios_tables_len);
1301         fw_cfg_add_file(s->fw_cfg, "etc/smbios/smbios-anchor",
1302                         smbios_anchor, smbios_anchor_len);
1303     }
1304 }
1305 
1306 static void virt_machine_done(Notifier *notifier, void *data)
1307 {
1308     RISCVVirtState *s = container_of(notifier, RISCVVirtState,
1309                                      machine_done);
1310     const MemMapEntry *memmap = virt_memmap;
1311     MachineState *machine = MACHINE(s);
1312     target_ulong start_addr = memmap[VIRT_DRAM].base;
1313     target_ulong firmware_end_addr, kernel_start_addr;
1314     const char *firmware_name = riscv_default_firmware_name(&s->soc[0]);
1315     uint64_t fdt_load_addr;
1316     uint64_t kernel_entry = 0;
1317     BlockBackend *pflash_blk0;
1318 
1319     /*
1320      * An user provided dtb must include everything, including
1321      * dynamic sysbus devices. Our FDT needs to be finalized.
1322      */
1323     if (machine->dtb == NULL) {
1324         finalize_fdt(s);
1325     }
1326 
1327     /*
1328      * Only direct boot kernel is currently supported for KVM VM,
1329      * so the "-bios" parameter is not supported when KVM is enabled.
1330      */
1331     if (kvm_enabled()) {
1332         if (machine->firmware) {
1333             if (strcmp(machine->firmware, "none")) {
1334                 error_report("Machine mode firmware is not supported in "
1335                              "combination with KVM.");
1336                 exit(1);
1337             }
1338         } else {
1339             machine->firmware = g_strdup("none");
1340         }
1341     }
1342 
1343     firmware_end_addr = riscv_find_and_load_firmware(machine, firmware_name,
1344                                                      start_addr, NULL);
1345 
1346     pflash_blk0 = pflash_cfi01_get_blk(s->flash[0]);
1347     if (pflash_blk0) {
1348         if (machine->firmware && !strcmp(machine->firmware, "none") &&
1349             !kvm_enabled()) {
1350             /*
1351              * Pflash was supplied but bios is none and not KVM guest,
1352              * let's overwrite the address we jump to after reset to
1353              * the base of the flash.
1354              */
1355             start_addr = virt_memmap[VIRT_FLASH].base;
1356         } else {
1357             /*
1358              * Pflash was supplied but either KVM guest or bios is not none.
1359              * In this case, base of the flash would contain S-mode payload.
1360              */
1361             riscv_setup_firmware_boot(machine);
1362             kernel_entry = virt_memmap[VIRT_FLASH].base;
1363         }
1364     }
1365 
1366     if (machine->kernel_filename && !kernel_entry) {
1367         kernel_start_addr = riscv_calc_kernel_start_addr(&s->soc[0],
1368                                                          firmware_end_addr);
1369 
1370         kernel_entry = riscv_load_kernel(machine, &s->soc[0],
1371                                          kernel_start_addr, true, NULL);
1372     }
1373 
1374     fdt_load_addr = riscv_compute_fdt_addr(memmap[VIRT_DRAM].base,
1375                                            memmap[VIRT_DRAM].size,
1376                                            machine);
1377     riscv_load_fdt(fdt_load_addr, machine->fdt);
1378 
1379     /* load the reset vector */
1380     riscv_setup_rom_reset_vec(machine, &s->soc[0], start_addr,
1381                               virt_memmap[VIRT_MROM].base,
1382                               virt_memmap[VIRT_MROM].size, kernel_entry,
1383                               fdt_load_addr);
1384 
1385     /*
1386      * Only direct boot kernel is currently supported for KVM VM,
1387      * So here setup kernel start address and fdt address.
1388      * TODO:Support firmware loading and integrate to TCG start
1389      */
1390     if (kvm_enabled()) {
1391         riscv_setup_direct_kernel(kernel_entry, fdt_load_addr);
1392     }
1393 
1394     virt_build_smbios(s);
1395 
1396     if (virt_is_acpi_enabled(s)) {
1397         virt_acpi_setup(s);
1398     }
1399 }
1400 
1401 static void virt_machine_init(MachineState *machine)
1402 {
1403     const MemMapEntry *memmap = virt_memmap;
1404     RISCVVirtState *s = RISCV_VIRT_MACHINE(machine);
1405     MemoryRegion *system_memory = get_system_memory();
1406     MemoryRegion *mask_rom = g_new(MemoryRegion, 1);
1407     DeviceState *mmio_irqchip, *virtio_irqchip, *pcie_irqchip;
1408     int i, base_hartid, hart_count;
1409     int socket_count = riscv_socket_count(machine);
1410 
1411     /* Check socket count limit */
1412     if (VIRT_SOCKETS_MAX < socket_count) {
1413         error_report("number of sockets/nodes should be less than %d",
1414             VIRT_SOCKETS_MAX);
1415         exit(1);
1416     }
1417 
1418     if (!virt_aclint_allowed() && s->have_aclint) {
1419         error_report("'aclint' is only available with TCG acceleration");
1420         exit(1);
1421     }
1422 
1423     /* Initialize sockets */
1424     mmio_irqchip = virtio_irqchip = pcie_irqchip = NULL;
1425     for (i = 0; i < socket_count; i++) {
1426         g_autofree char *soc_name = g_strdup_printf("soc%d", i);
1427 
1428         if (!riscv_socket_check_hartids(machine, i)) {
1429             error_report("discontinuous hartids in socket%d", i);
1430             exit(1);
1431         }
1432 
1433         base_hartid = riscv_socket_first_hartid(machine, i);
1434         if (base_hartid < 0) {
1435             error_report("can't find hartid base for socket%d", i);
1436             exit(1);
1437         }
1438 
1439         hart_count = riscv_socket_hart_count(machine, i);
1440         if (hart_count < 0) {
1441             error_report("can't find hart count for socket%d", i);
1442             exit(1);
1443         }
1444 
1445         object_initialize_child(OBJECT(machine), soc_name, &s->soc[i],
1446                                 TYPE_RISCV_HART_ARRAY);
1447         object_property_set_str(OBJECT(&s->soc[i]), "cpu-type",
1448                                 machine->cpu_type, &error_abort);
1449         object_property_set_int(OBJECT(&s->soc[i]), "hartid-base",
1450                                 base_hartid, &error_abort);
1451         object_property_set_int(OBJECT(&s->soc[i]), "num-harts",
1452                                 hart_count, &error_abort);
1453         sysbus_realize(SYS_BUS_DEVICE(&s->soc[i]), &error_fatal);
1454 
1455         if (virt_aclint_allowed() && s->have_aclint) {
1456             if (s->aia_type == VIRT_AIA_TYPE_APLIC_IMSIC) {
1457                 /* Per-socket ACLINT MTIMER */
1458                 riscv_aclint_mtimer_create(memmap[VIRT_CLINT].base +
1459                             i * RISCV_ACLINT_DEFAULT_MTIMER_SIZE,
1460                         RISCV_ACLINT_DEFAULT_MTIMER_SIZE,
1461                         base_hartid, hart_count,
1462                         RISCV_ACLINT_DEFAULT_MTIMECMP,
1463                         RISCV_ACLINT_DEFAULT_MTIME,
1464                         RISCV_ACLINT_DEFAULT_TIMEBASE_FREQ, true);
1465             } else {
1466                 /* Per-socket ACLINT MSWI, MTIMER, and SSWI */
1467                 riscv_aclint_swi_create(memmap[VIRT_CLINT].base +
1468                             i * memmap[VIRT_CLINT].size,
1469                         base_hartid, hart_count, false);
1470                 riscv_aclint_mtimer_create(memmap[VIRT_CLINT].base +
1471                             i * memmap[VIRT_CLINT].size +
1472                             RISCV_ACLINT_SWI_SIZE,
1473                         RISCV_ACLINT_DEFAULT_MTIMER_SIZE,
1474                         base_hartid, hart_count,
1475                         RISCV_ACLINT_DEFAULT_MTIMECMP,
1476                         RISCV_ACLINT_DEFAULT_MTIME,
1477                         RISCV_ACLINT_DEFAULT_TIMEBASE_FREQ, true);
1478                 riscv_aclint_swi_create(memmap[VIRT_ACLINT_SSWI].base +
1479                             i * memmap[VIRT_ACLINT_SSWI].size,
1480                         base_hartid, hart_count, true);
1481             }
1482         } else if (tcg_enabled()) {
1483             /* Per-socket SiFive CLINT */
1484             riscv_aclint_swi_create(
1485                     memmap[VIRT_CLINT].base + i * memmap[VIRT_CLINT].size,
1486                     base_hartid, hart_count, false);
1487             riscv_aclint_mtimer_create(memmap[VIRT_CLINT].base +
1488                         i * memmap[VIRT_CLINT].size + RISCV_ACLINT_SWI_SIZE,
1489                     RISCV_ACLINT_DEFAULT_MTIMER_SIZE, base_hartid, hart_count,
1490                     RISCV_ACLINT_DEFAULT_MTIMECMP, RISCV_ACLINT_DEFAULT_MTIME,
1491                     RISCV_ACLINT_DEFAULT_TIMEBASE_FREQ, true);
1492         }
1493 
1494         /* Per-socket interrupt controller */
1495         if (s->aia_type == VIRT_AIA_TYPE_NONE) {
1496             s->irqchip[i] = virt_create_plic(memmap, i,
1497                                              base_hartid, hart_count);
1498         } else {
1499             s->irqchip[i] = virt_create_aia(s->aia_type, s->aia_guests,
1500                                             memmap, i, base_hartid,
1501                                             hart_count);
1502         }
1503 
1504         /* Try to use different IRQCHIP instance based device type */
1505         if (i == 0) {
1506             mmio_irqchip = s->irqchip[i];
1507             virtio_irqchip = s->irqchip[i];
1508             pcie_irqchip = s->irqchip[i];
1509         }
1510         if (i == 1) {
1511             virtio_irqchip = s->irqchip[i];
1512             pcie_irqchip = s->irqchip[i];
1513         }
1514         if (i == 2) {
1515             pcie_irqchip = s->irqchip[i];
1516         }
1517     }
1518 
1519     if (kvm_enabled() && virt_use_kvm_aia(s)) {
1520         kvm_riscv_aia_create(machine, IMSIC_MMIO_GROUP_MIN_SHIFT,
1521                              VIRT_IRQCHIP_NUM_SOURCES, VIRT_IRQCHIP_NUM_MSIS,
1522                              memmap[VIRT_APLIC_S].base,
1523                              memmap[VIRT_IMSIC_S].base,
1524                              s->aia_guests);
1525     }
1526 
1527     if (riscv_is_32bit(&s->soc[0])) {
1528 #if HOST_LONG_BITS == 64
1529         /* limit RAM size in a 32-bit system */
1530         if (machine->ram_size > 10 * GiB) {
1531             machine->ram_size = 10 * GiB;
1532             error_report("Limiting RAM size to 10 GiB");
1533         }
1534 #endif
1535         virt_high_pcie_memmap.base = VIRT32_HIGH_PCIE_MMIO_BASE;
1536         virt_high_pcie_memmap.size = VIRT32_HIGH_PCIE_MMIO_SIZE;
1537     } else {
1538         virt_high_pcie_memmap.size = VIRT64_HIGH_PCIE_MMIO_SIZE;
1539         virt_high_pcie_memmap.base = memmap[VIRT_DRAM].base + machine->ram_size;
1540         virt_high_pcie_memmap.base =
1541             ROUND_UP(virt_high_pcie_memmap.base, virt_high_pcie_memmap.size);
1542     }
1543 
1544     s->memmap = virt_memmap;
1545 
1546     /* register system main memory (actual RAM) */
1547     memory_region_add_subregion(system_memory, memmap[VIRT_DRAM].base,
1548         machine->ram);
1549 
1550     /* boot rom */
1551     memory_region_init_rom(mask_rom, NULL, "riscv_virt_board.mrom",
1552                            memmap[VIRT_MROM].size, &error_fatal);
1553     memory_region_add_subregion(system_memory, memmap[VIRT_MROM].base,
1554                                 mask_rom);
1555 
1556     /*
1557      * Init fw_cfg. Must be done before riscv_load_fdt, otherwise the
1558      * device tree cannot be altered and we get FDT_ERR_NOSPACE.
1559      */
1560     s->fw_cfg = create_fw_cfg(machine);
1561     rom_set_fw(s->fw_cfg);
1562 
1563     /* SiFive Test MMIO device */
1564     sifive_test_create(memmap[VIRT_TEST].base);
1565 
1566     /* VirtIO MMIO devices */
1567     for (i = 0; i < VIRTIO_COUNT; i++) {
1568         sysbus_create_simple("virtio-mmio",
1569             memmap[VIRT_VIRTIO].base + i * memmap[VIRT_VIRTIO].size,
1570             qdev_get_gpio_in(virtio_irqchip, VIRTIO_IRQ + i));
1571     }
1572 
1573     gpex_pcie_init(system_memory, pcie_irqchip, s);
1574 
1575     create_platform_bus(s, mmio_irqchip);
1576 
1577     serial_mm_init(system_memory, memmap[VIRT_UART0].base,
1578         0, qdev_get_gpio_in(mmio_irqchip, UART0_IRQ), 399193,
1579         serial_hd(0), DEVICE_LITTLE_ENDIAN);
1580 
1581     sysbus_create_simple("goldfish_rtc", memmap[VIRT_RTC].base,
1582         qdev_get_gpio_in(mmio_irqchip, RTC_IRQ));
1583 
1584     for (i = 0; i < ARRAY_SIZE(s->flash); i++) {
1585         /* Map legacy -drive if=pflash to machine properties */
1586         pflash_cfi01_legacy_drive(s->flash[i],
1587                                   drive_get(IF_PFLASH, 0, i));
1588     }
1589     virt_flash_map(s, system_memory);
1590 
1591     /* load/create device tree */
1592     if (machine->dtb) {
1593         machine->fdt = load_device_tree(machine->dtb, &s->fdt_size);
1594         if (!machine->fdt) {
1595             error_report("load_device_tree() failed");
1596             exit(1);
1597         }
1598     } else {
1599         create_fdt(s, memmap);
1600     }
1601 
1602     s->machine_done.notify = virt_machine_done;
1603     qemu_add_machine_init_done_notifier(&s->machine_done);
1604 }
1605 
1606 static void virt_machine_instance_init(Object *obj)
1607 {
1608     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1609 
1610     virt_flash_create(s);
1611 
1612     s->oem_id = g_strndup(ACPI_BUILD_APPNAME6, 6);
1613     s->oem_table_id = g_strndup(ACPI_BUILD_APPNAME8, 8);
1614     s->acpi = ON_OFF_AUTO_AUTO;
1615 }
1616 
1617 static char *virt_get_aia_guests(Object *obj, Error **errp)
1618 {
1619     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1620 
1621     return g_strdup_printf("%d", s->aia_guests);
1622 }
1623 
1624 static void virt_set_aia_guests(Object *obj, const char *val, Error **errp)
1625 {
1626     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1627 
1628     s->aia_guests = atoi(val);
1629     if (s->aia_guests < 0 || s->aia_guests > VIRT_IRQCHIP_MAX_GUESTS) {
1630         error_setg(errp, "Invalid number of AIA IMSIC guests");
1631         error_append_hint(errp, "Valid values be between 0 and %d.\n",
1632                           VIRT_IRQCHIP_MAX_GUESTS);
1633     }
1634 }
1635 
1636 static char *virt_get_aia(Object *obj, Error **errp)
1637 {
1638     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1639     const char *val;
1640 
1641     switch (s->aia_type) {
1642     case VIRT_AIA_TYPE_APLIC:
1643         val = "aplic";
1644         break;
1645     case VIRT_AIA_TYPE_APLIC_IMSIC:
1646         val = "aplic-imsic";
1647         break;
1648     default:
1649         val = "none";
1650         break;
1651     };
1652 
1653     return g_strdup(val);
1654 }
1655 
1656 static void virt_set_aia(Object *obj, const char *val, Error **errp)
1657 {
1658     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1659 
1660     if (!strcmp(val, "none")) {
1661         s->aia_type = VIRT_AIA_TYPE_NONE;
1662     } else if (!strcmp(val, "aplic")) {
1663         s->aia_type = VIRT_AIA_TYPE_APLIC;
1664     } else if (!strcmp(val, "aplic-imsic")) {
1665         s->aia_type = VIRT_AIA_TYPE_APLIC_IMSIC;
1666     } else {
1667         error_setg(errp, "Invalid AIA interrupt controller type");
1668         error_append_hint(errp, "Valid values are none, aplic, and "
1669                           "aplic-imsic.\n");
1670     }
1671 }
1672 
1673 static bool virt_get_aclint(Object *obj, Error **errp)
1674 {
1675     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1676 
1677     return s->have_aclint;
1678 }
1679 
1680 static void virt_set_aclint(Object *obj, bool value, Error **errp)
1681 {
1682     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1683 
1684     s->have_aclint = value;
1685 }
1686 
1687 bool virt_is_acpi_enabled(RISCVVirtState *s)
1688 {
1689     return s->acpi != ON_OFF_AUTO_OFF;
1690 }
1691 
1692 static void virt_get_acpi(Object *obj, Visitor *v, const char *name,
1693                           void *opaque, Error **errp)
1694 {
1695     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1696     OnOffAuto acpi = s->acpi;
1697 
1698     visit_type_OnOffAuto(v, name, &acpi, errp);
1699 }
1700 
1701 static void virt_set_acpi(Object *obj, Visitor *v, const char *name,
1702                           void *opaque, Error **errp)
1703 {
1704     RISCVVirtState *s = RISCV_VIRT_MACHINE(obj);
1705 
1706     visit_type_OnOffAuto(v, name, &s->acpi, errp);
1707 }
1708 
1709 static HotplugHandler *virt_machine_get_hotplug_handler(MachineState *machine,
1710                                                         DeviceState *dev)
1711 {
1712     MachineClass *mc = MACHINE_GET_CLASS(machine);
1713 
1714     if (device_is_dynamic_sysbus(mc, dev) ||
1715         object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
1716         return HOTPLUG_HANDLER(machine);
1717     }
1718     return NULL;
1719 }
1720 
1721 static void virt_machine_device_plug_cb(HotplugHandler *hotplug_dev,
1722                                         DeviceState *dev, Error **errp)
1723 {
1724     RISCVVirtState *s = RISCV_VIRT_MACHINE(hotplug_dev);
1725 
1726     if (s->platform_bus_dev) {
1727         MachineClass *mc = MACHINE_GET_CLASS(s);
1728 
1729         if (device_is_dynamic_sysbus(mc, dev)) {
1730             platform_bus_link_device(PLATFORM_BUS_DEVICE(s->platform_bus_dev),
1731                                      SYS_BUS_DEVICE(dev));
1732         }
1733     }
1734 
1735     if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
1736         create_fdt_virtio_iommu(s, pci_get_bdf(PCI_DEVICE(dev)));
1737     }
1738 }
1739 
1740 static void virt_machine_class_init(ObjectClass *oc, void *data)
1741 {
1742     MachineClass *mc = MACHINE_CLASS(oc);
1743     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
1744 
1745     mc->desc = "RISC-V VirtIO board";
1746     mc->init = virt_machine_init;
1747     mc->max_cpus = VIRT_CPUS_MAX;
1748     mc->default_cpu_type = TYPE_RISCV_CPU_BASE;
1749     mc->pci_allow_0_address = true;
1750     mc->possible_cpu_arch_ids = riscv_numa_possible_cpu_arch_ids;
1751     mc->cpu_index_to_instance_props = riscv_numa_cpu_index_to_props;
1752     mc->get_default_cpu_node_id = riscv_numa_get_default_cpu_node_id;
1753     mc->numa_mem_supported = true;
1754     /* platform instead of architectural choice */
1755     mc->cpu_cluster_has_numa_boundary = true;
1756     mc->default_ram_id = "riscv_virt_board.ram";
1757     assert(!mc->get_hotplug_handler);
1758     mc->get_hotplug_handler = virt_machine_get_hotplug_handler;
1759 
1760     hc->plug = virt_machine_device_plug_cb;
1761 
1762     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_RAMFB_DEVICE);
1763 #ifdef CONFIG_TPM
1764     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_TPM_TIS_SYSBUS);
1765 #endif
1766 
1767     object_class_property_add_bool(oc, "aclint", virt_get_aclint,
1768                                    virt_set_aclint);
1769     object_class_property_set_description(oc, "aclint",
1770                                           "(TCG only) Set on/off to "
1771                                           "enable/disable emulating "
1772                                           "ACLINT devices");
1773 
1774     object_class_property_add_str(oc, "aia", virt_get_aia,
1775                                   virt_set_aia);
1776     object_class_property_set_description(oc, "aia",
1777                                           "Set type of AIA interrupt "
1778                                           "controller. Valid values are "
1779                                           "none, aplic, and aplic-imsic.");
1780 
1781     object_class_property_add_str(oc, "aia-guests",
1782                                   virt_get_aia_guests,
1783                                   virt_set_aia_guests);
1784     {
1785         g_autofree char *str =
1786             g_strdup_printf("Set number of guest MMIO pages for AIA IMSIC. "
1787                             "Valid value should be between 0 and %d.",
1788                             VIRT_IRQCHIP_MAX_GUESTS);
1789         object_class_property_set_description(oc, "aia-guests", str);
1790     }
1791 
1792     object_class_property_add(oc, "acpi", "OnOffAuto",
1793                               virt_get_acpi, virt_set_acpi,
1794                               NULL, NULL);
1795     object_class_property_set_description(oc, "acpi",
1796                                           "Enable ACPI");
1797 }
1798 
1799 static const TypeInfo virt_machine_typeinfo = {
1800     .name       = MACHINE_TYPE_NAME("virt"),
1801     .parent     = TYPE_MACHINE,
1802     .class_init = virt_machine_class_init,
1803     .instance_init = virt_machine_instance_init,
1804     .instance_size = sizeof(RISCVVirtState),
1805     .interfaces = (InterfaceInfo[]) {
1806          { TYPE_HOTPLUG_HANDLER },
1807          { }
1808     },
1809 };
1810 
1811 static void virt_machine_init_register_types(void)
1812 {
1813     type_register_static(&virt_machine_typeinfo);
1814 }
1815 
1816 type_init(virt_machine_init_register_types)
1817